Performance of rank-minimization under different scenarios: a simulation study focusing on baseline covariates imbalances in clinical trials

摘要

Clinical trials are often considered to be the gold standard for assessing effectiveness and safety of medical treatments and public health interventions. The validity of inferences from clinical trials depends on randomizing subjects to different treatment groups. Although simple randomization is the most common approach, and generally prevents differences in baseline covariate imbalances between groups, other approaches may be necessary for balancing covariate distributions within important strata. However, the performance of stratified randomization may be limited when the sample size is small and there are many strata. These scenarios may be better addressed through minimization, or rank-minimization algorithms. udThe concept of rank-minimization is straightforward but very little research has been published on the topic. To address this gap in the literature, we conducted a simulation study to investigate how rank-minimization performed, compared to Taves’ minimization, with different sample sizes and baseline covariate distributions. udResults indicated that both sample size and covariate distributions influence the performance of rank-minimization and minimization. Overall, rank-minimization yields better properties, and larger sample sizes yield better properties for both methods. However, the performance for both methods decreases when the distribution is more skewed. Results of this study provide researchers with more information to decide between randomization methods for their specific applications.udPublic Health Significance: In clinical trials, the comparability of subjects between different treatment groups is critical to validity of the subsequent inferences. Since clinical trials are often considered the gold standard for assessing medical treatments and public health interventions, and the trials are usually expensive and time-consuming to conduct, optimizing the randomization process represents a highly significant aspect of public health research. Consulting results of the simulation study will provide additional information for researchers to decide the best method for randomization for different size data sets and different covariate distributions encountered in practice.